Empirical Mode Decomposition Filtering of Wind Profiles

Abstract

A filtering method based on the empirical mode decomposition (EMD) is developed to remove spurious, quasi-periodic features in measured wind profiles. These oscillatory features are associated with the pendulum motion of the sonde responding to the balloon’s lateral self-induced motions. Since these oscillations are not indicative of the actual wind, current balloon sounding systems apply digital filters to remove these artifacts. Data from two GPS-based wind profiling systems, the National Weather Service Radiosonde Replacement System (RRS) and the Low-Resolution Automated Meteorological Profiling System (AMPS), will be examined. These profiling systems employ pre-defined low-pass digital filters that have prescribed frequency cutoffs. Since the frequencies of the oscillations vary both temporally and spatially, conservative filter cutoffs must be used to ensure that these artifacts are suppressed. On the other hand, the EMD is a data-driven approach that resolves a time series as a summation of its unique intrinsic mode functions (IMF). As such, the EMD-based filtering is able to adaptively identify and remove the IMFs associated with these oscillations. The filter’s characteristics will be examined using synthetic data that exhibits the random oscillations of a pendulum responding to fractional Brownian motions. The practical effectiveness of the EMD filtering will be demonstrated using the RRS and AMPS data.